A Survey on Long-Tailed Visual Recognition

The heavy reliance on data is one of the major reasons that currently limit the development of deep learning. Data quality directly dominates the effect of deep learning models, and the long-tailed distribution is one of the factors affecting data quality. The long-tailed phenomenon is prevalent due to the prevalence of power law in nature. In this case, the performance of deep learning models is often dominated by the head classes while the learning of the tail classes is severely underdeveloped. In order to learn adequately for all classes, many researchers have studied and preliminarily addressed the long-tailed problem. In this survey, we focus on the problems caused by long-tailed data distribution, sort out the representative long-tailed visual recognition datasets and summarize some mainstream long-tailed studies. Specifically, we summarize these studies into ten categories from the perspective of representation learning, and outline the highlights and limitations of each category. Besides, we have studied four quantitative metrics for evaluating the imbalance, and suggest using the Gini coefficient to evaluate the long-tailedness of a dataset. Based on the Gini coefficient, we quantitatively study 20 widely-used and large-scale visual datasets proposed in the last decade, and find that the long-tailed phenomenon is widespread and has not been fully studied. Finally, we provide several future directions for the development of long-tailed learning to provide more ideas for readers.

     

Designing Metallic and Insulating Nanocrystal Heterostructures to Fabricate Highly Sensitive and Solution Processed Strain Gauges for Wearable Sensors

All‐solution processed, high‐performance wearable strain sensors are demonstrated using heterostructure nanocrystal (NC) solids. By incorporating insulating artificial atoms of CdSe quantum dot NCs into metallic artificial atoms of Au NC thin film matrix, metal–insulator heterostructures are designed. This hybrid structure results in a shift close to the percolation threshold, modifying the charge transport mechanism and enhancing sensitivity in accordance with the site percolation theory. The number of electrical pathways is also manipulated by creating nanocracks to further increase its sensitivity, inspired from the bond percolation theory. The combination of the two strategies achieves gauge factor up to 5045, the highest sensitivity recorded among NC‐based strain gauges. These strain sensors show high reliability, durability, frequency stability, and negligible hysteresis. The fundamental charge transport behavior of these NC solids is investigated and the combined site and bond percolation theory is developed to illuminate the origin of their enhanced sensitivity. Finally, all NC‐based and solution‐processed strain gauge sensor arrays are fabricated, which effectively measure the motion of each finger joint, the pulse of heart rate, and the movement of vocal cords of human. This work provides a pathway for designing low‐cost and high‐performance electronic skin or wearable devices.     

数控机床操控器人机界面优化设计研究

目的控制器和显示器共同组成数控机床控制面板的整体界面,是机床接受与执行命令的交互平台。从人机交互角度,聚焦于控制器的界面布局进行优化设计。方法基于人机工程学理论的优化流程,采用现场访谈、现场观察等方法,将人机工程学的理论转化为具体考核指标。结论从人机工程学的研究角度出发,结合工业设计、用户体验设计学等内容,从功能区域分区、操纵器色彩设计等方面优化数控机床控制器的界面,提高系统整体效能及使用者满意度,为数控机床系统开发设计提供参考。     

Localization of atrial natriuretic peptide receptors in the rat tongue and hard palate

The effect of a naturally occurring plant phenolic constituent (the acylphloroglucinol derivative, jensenone, derived from Eucalyptus jensenii) on the food intake of two folivorous marsupials, the common ringtail (Pseudocheirus peregrinus) and the common brushtail possum (Trichosurus vulpecula) was studied. When fed diets containing varying concentrations of jensenone, both species regulated their intake of jensenone so as not to exceed a ceiling intake. This ceiling was about twice as high for common ringtails as for common brushtails from northern Australia. Southern populations of common ringtails showed greatly reduced capacities to tolerate jensenone. When common brushtails were injected (0.5 mg.kg-0.75 body mass) with ondansetron (a selective antagonist of serotonin 5HT3 receptors), they ate significantly more jensenone than animals injected with physiological saline. The same pattern was observed when common ringtails were fed diets containing both jensenone and ondansetron (0.0035 mg.g-1 wet mass of diet). Ondansetron injection had no effect on food intake when the food did not contain jensenone while the addition of higher doses of ondansetron to diets of common ringtails very slightly reduced food intakes of a non-jensenone diet. When common brushtails were given 50 mg of jensenone by gastric lavage, their average subsequent intake of dietary jensenone matched the difference between the daily threshold and the dose given, although the response of individuals was highly variable. Lavage with water alone had no effect on subsequent jensenone intake compared with the pre-dose period. We interpret these results as evidence that the antifeedant effects of jensenone and related compounds are partly mediated by serotonin action on 5HT3 receptors most likely via "nausea" to condition a food aversion.     

Absorption effect in the calculation of a second-order nonlinear coefficient from the data of a maker fringe experiment

The effect of light absorption by sample in the analysis of Makerfringe data for estimating a second-order nonlinear coefficient hasbeen studied experimentally. Two theories, one by Jerphagnon andKurtz that neglects the absorption effect and one by Herman and Haydenthat takes into account the absorption effect, were compared with theexperimental results. It was found that Jerphagnon and Kurtz'sformula was unable to predict correctly not only the magnitude but alsothe incident angle dependence or the sample thickness dependence of thesecond harmonic signal generated by the sample with strong absorption, whereas the theory by Herman and Hayden was able to make thosepredictions fairly well. It was also found that the error in theestimated nonlinear coefficient when one uses Jerphagnon and Kurtz'sformula could be as large as 2-4 times the true value, depending onsample thickness.     

Optical inhomogeneity and microstructure of ZrO(2) thin films prepared by ion-assisted deposition

The index inhomogeneity and the microstructure of ZrO(2) films prepared by Ar-ion-assisted deposition are investigated. The results show that as the Ar-ion momentum transferred to the growing film increases, the average refractive index increases, the vacuum-to-air spectral shift becomes almost zero, the sign of relative inhomogeneity transits from negative to positive, and the void fraction of the top layer next to air becomes smaller than that of the bottom one. These optical properties result from the improved packing density and denser outer region next to air. The Ar-ion bombardment also induces the changes in microstructure of ZrO(2) films, such as the preferential (111) orientation of cubic phase, increase in compressive stress, and reduction of surface roughness.